Metareasoning Based Self Adaptive Tracking

In this paper we describe a system that tracks vehicles from overhead video using a self-adaptive bank of Kalman filters. The system utilizes a bank of base-level reasoners that promote their own hypotheses about vehicle models and make predictions about future vehicle motion. By evaluating how well the base reasoners predictions are realized by the vehicles, metareasoning allows leading base reasoners to be selected and modified in the course of the passage of a vehicle through the video. It is shown how multiple hypothesis tracking within a self-adaptive framework produces superior object tracking and prediction in the face of noisy data.